Founders Should Chase Problems, Not Entrepreneurship: A Conversation with Sam Larson, Founder & CEO of Hone
By Xiao He
Introduction
When we first met at an AI eval event, Sam struck me as calm, technical, and deeply thoughtful. What I didn’t expect was the winding path behind that presence: two years in Korea, studying econometrics instead of computer science, a stint in private equity that led to burnout, an almost-enlistment in the Navy, and eventually, building an AI infrastructure startup.
Sam is now the founder of Hone, a platform that helps companies monitor, evaluate, and continuously improve AI agents in production, especially for non-technical teams who actually use them day to day.
The Interview
Xiao He
Thank you so much for taking the time to chat today. Could you start by introducing yourself to our readers?
Sam Larson
Yeah. I’ll say a little bit about my story.
I’m Sam. I’m originally from New York and grew up in the suburbs. After high school, I actually lived in Korea for two years — kind of gap years — and then went to school where I studied econometrics. So I didn’t study computer science. It was more statistics and theoretical statistics.
We had some coursework using machine learning, but looking back now it feels very old-school: regularized regressions, classic NLP methods, random forests. Neural nets were considered cutting-edge at the time. Now everything is transformers and it’s a whole new world.
After that, I went into private equity in New York. And honestly, I was hating my life. I stayed about a year, then quit without really having a plan. While quitting, I applied to Georgia Tech’s online Master’s in Computer Science with a focus on machine learning so I could stay in New York.
I got a job at a startup called Musetax, building a financial advisor product for people who normally wouldn’t have access to tax or financial advice — mostly minority groups. Around that time, I was also debating joining the Navy. I didn’t end up going because of some health issues.
But during all of that, I realized: I actually really like this AI stuff.
So I leaned into it.
Eventually, I moved to Silicon Valley and became a more serious AI engineer. That’s where I really learned software fundamentals — architecture, engineering practices — combined with my data science background. That combo has been really powerful for me.
All of that led me to where I am now, building Hone.
Hone essentially does three things: it monitors, evaluates, and iterates on AI agents in production.
You can think of competitors like LangSmith or Braintrust, but we’re differentiated. We try to appeal to less technical users — the domain experts — because they’re the ones who really understand how the agents should behave. So we built something they can use directly.
We also maintain versioned prompts, tests, and benchmarks built from production logs. As you iterate, you always have a gold standard.
A lot of enterprises struggle here. There was an MIT study that showed 95% of enterprise AI projects fail. Maintainability is one of the biggest reasons. Once something is live, nobody knows how to keep it good.
That’s exactly the problem we’re trying to solve.
I also do consulting for another company called Avara. They build AI avatar agents. They had to deploy thousands of agents across different organizations, and the managers using them weren’t technical.
They needed rapid iteration and quality assurance without engineers constantly stepping in.
So Hone was actually born because we needed it internally first. They became our first customer.
Xiao
How did you realize this pain point could become its own company, not just an internal tool?
Sam
Two reasons.
First, I didn’t think we could even build it properly inside the organization because of legacy systems and bureaucracy. Prompts were scattered in code everywhere. It’s messy. Hard to track. Hard to version.
Second, if we had this problem, other companies definitely had it too.
I tried all the major tools — Braintrust, LangSmith, Galileo — and none really worked for us. We needed something more automated and something that could extract knowledge from domain experts easily.
That’s actually why we use voice agents. You can’t just sit someone in front of a form and expect them to type everything they know. You need them to talk it out, like we’re doing now.
Xiao
Is voice input your biggest differentiator?
Sam
Originally, yes. I thought voice would be the main differentiator.
And it is helpful. The voice agents act like thought partners and ask discovery questions.
But what surprised us is that something else became huge: prompt management and versioning.
Taking prompts out of the engineering loop and letting non-technical domain experts manage them themselves turned out to be incredibly valuable.
Engineers love it because they don’t have to do constant PRs for tiny wording changes. Product managers love it because they can update instantly.
It’s kind of funny — we built it because it felt intuitive, not because we thought it was the killer feature. But customers told us it was.
That’s been a big lesson.
Xiao
How did you discover that shift?
Sam
It wasn’t some big pivot. The functionality was already there.
We just noticed engineers kept reacting strongly to that feature. So in conversations, we started emphasizing it more.
It shaped the messaging naturally.
It’s still recent — like the last couple weeks — so we haven’t changed the website copy or anything. But that’s how you learn. You pay attention to what people light up about.
Xiao
How did you find your first customers beyond Avara?
Sam
My co-founders and I are all AI engineers, so the first three customers were basically our own companies.
One is a major university healthcare system using AI agents for patient intake and support. We sit between the agents and the nurses who deal with the output.
Another is my company.
And another co-founder implemented it across his clients.
Now we’re also working with AI consulting firms who build agents for clients but don’t want to maintain them forever. Hone helps transfer that maintenance back to the client.
Xiao
How did you and your co-founders meet?
Sam
One of them and I knew each other from college. He’s kind of the “glue guy.” He knew the third co-founder from growing up together.
I came to him and said, “I’ve got this idea. I think it could be big.”
We started building an MVP. It felt real. Then we brought the third guy in, who’s an excellent backend engineer.
And now we’re just moving fast and having fun.
Xiao
You started with real customers instead of raising first. That’s good and rare!
Sam
Yeah, I think it’s important. Raising money without product-market fit doesn’t give you leverage. You just have money. Customers are a leverage.
Money is just a tool to get customers. Don’t raise too fast or too much.
Xiao
Will you raise eventually?
Sam
We plan to bootstrap until we have about ten happy, consistently renewing customers and a certain ARR number.
Then we’d raise mostly for marketing. In the AI era, you don’t need a huge engineering team. You just need strong co-founders.
Xiao
What’s the plan for the next six months?
Sam
More customers. Product hardening. Exploring this prompt-management niche. Testing hypotheses around consultants and forward-deployed teams.
Then once the message is clear, we scale marketing.
Xiao
Did you always want to be an entrepreneur?
Sam
No, and I don’t think people should set out to be entrepreneurs.
You should set out to solve a problem. If solving that problem requires building a company, then that’s what you do.
Entrepreneurship shouldn’t be “I want to work for myself.” It should be “I care about this problem and I think I can solve it.”
Xiao
At the end of every interview, we ask for a recommendation — a book, film, or podcast.
Sam
I’d say One Hundred Years of Solitude.
It’s unlike most American novels. It’s not about one person. It’s about a family and patterns over time. It expands your perspective beyond the individual ego.
It’s beautifully written and just a totally different way to experience storytelling.
Highly recommend.